Abstract

Microaneurysms, tiny, circular red dots that occur in retinal fundus images, are one of the earliest symptoms of diabetic retinopathy. Because microaneurysms are small and delicate, detecting them can be difficult. Their small size and cunning character make automatic detection of them difficult. The automatic detection of microaneurysms in retinal fundus images is proposed in this research using a local Fourier transform and neighbourhood analysis-based multi-scale approach technique. The suggested method is broken down into three stages: image preprocessing, the detection of retinal vessels and microaneurysm candidates, and labelling of the candidates. A multi-scale framework is used to develop every stage of the algorithm, with the exception of the initial image preprocessing, giving the mechanism for efficient microaneurysms detection. In contrast to the short-time Fourier transform, which extracts the neighbourhood of each pixel and calculates each local Fourier transform separately, the local Fourier transform is employed in this study to extract the MA. After that, neighbourhood analysis is performed to name the microaneurysm because the item is actually a collection of independent little images rather than the entire image. Three separate data sets and different types of performance indicators are used to examine the robustness of the proposed model. Through the prominent performance, the proposed model is able to outperform other existing models. The classification accuracy of the proposed method for MESSIDOR and ORIGA data set is 99.28% and 98.95%, respectively.

中文翻译：

---

使用局部傅里叶变换和邻域分析检测眼底图像中的微动脉瘤

摘要

微动脉瘤，即视网膜眼底图像中出现的微小圆形红点，是糖尿病视网膜病变的最早症状之一。由于微动脉瘤小且脆弱，因此检测它们可能很困难。它们的小尺寸和狡猾的特性使得自动检测它们变得困难。本研究提出使用局部傅里叶变换和基于邻域分析的多尺度方法技术自动检测视网膜眼底图像中的微动脉瘤。建议的方法分为三个阶段：图像预处理、视网膜血管和微动脉瘤候选者的检测以及候选者的标记。多尺度框架用于开发算法的每个阶段（初始图像预处理除外），从而提供有效的微动脉瘤检测机制。与短时傅里叶变换提取每个像素的邻域并分别计算每个局部傅里叶变换不同，本研究采用局部傅里叶变换来提取MA。之后，进行邻域分析来命名微动脉瘤，因为该项目实际上是独立小图像的集合而不是整个图像。使用三个独立的数据集和不同类型的性能指标来检查所提出模型的稳健性。通过突出的性能，所提出的模型能够优于其他现有模型。该方法对MESSIDOR和ORIGA数据集的分类准确率分别为99.28%和98.95%。